From Performers to Creators: Understanding Retired Women's Perceptions of Technology-Enhanced Dance Performance

Zheng and colleagues study how interactive-dance and generative-AI technologies can be designed around the needs of retired women dancers in China — a population estimated at over 100 million, for whom community dance is a major post-retirement practice but where stage production sits behind a wall of professional equipment, technical vocabulary, and age-related physical and digital-literacy barriers. The paper takes a two-workshop, research-probe approach. Workshop I (15 dancers, ages 51-67) is an exploratory, four-phase session combining questionnaire/interview, an introduction to a taxonomy of interactive dance (movement-, physiology-, or prop-driven inputs mapped to geometric, humanoid, or scene outputs), hands-on probes spanning all six cells of that taxonomy, and a co-design brainstorm over the participants' own past performance videos. From Workshop I the team extracts three design considerations — low-barrier entry, embodied movement-visual coupling, and creative ownership — and translates them into StageTailor, a two-step probe that pairs an LLM (Ernie Bot) expanding keyword input into a scene description, a text-to-video model (Haiper) rendering that scene as a backdrop, and a Kinect/IMU motion-capture layer that overlays motion-responsive visual effects. Workshop II (16 dancers, 52-67, seven returning) deploys StageTailor in a 100-minute hands-on evaluation with 5-point Likert ratings, observation, and semi-structured focus groups analysed via thematic analysis. The framing is explicitly age-sensitive creative AI mediation, drawing on participatory-design scholarship and the COM-B behaviour-change model (capability, opportunity, motivation) to argue that AI should scaffold rather than substitute for older dancers' embodied expertise.

Workshop I surfaced three persistent frustrations with current community-stage practice: backdrops are thematically disconnected from the dance (cited by 13 of 15 participants), dancers lack the time or technical skill to build appropriate visuals (N=8), and repeated use of the same backdrop over months produces fatigue and disengagement. Nearly all participants (N=14) preferred pre-designing visuals rather than improvising live because of the cognitive load of group synchrony. StageTailor in Workshop II achieved an overall satisfaction rating of 4.07/5. Keyword-based LLM input was strongly preferred to full-scene writing (usability 3.88/5, creativity-stimulation 4.00/5), with 14 of 16 participants choosing keywords and iterating a mean of 2.81 times per person. Generated videos produced high initial excitement (4+/5) but satisfaction dropped to 3.33/5 on close inspection as participants noticed compositional, temporal, and stylistic mismatches with their embodied vision (mean 3.25 refinements per group). Motion-responsive visual effects previewed at 4.0/5 but fell to 3.0/5 once composited, mainly because effects felt "pasted on" rather than integrated with the scene semantics. Despite these gaps, participants unanimously reported a role shift — explicitly described with phrases like "finally lets me create something that feels like mine" and "like performing somewhere professional, even with just our community's single screen" — and a willingness-to-participate-in-stage-creation score of 3.88/5. The authors distil eight design implications covering reflective keyword input, visual selectors for abstract attributes, multi-output reflection, emotion- and time-structured scaffolding, genre-aligned motion mapping, scene-grounded effects, visible authorship attribution, and collaborative/community-based co-creation.

For accessibility practitioners, the paper is a useful reminder that "accessibility" extends beyond assistive technology for disability into age-sensitive creative tooling, where the relevant barriers are digital literacy, vocabulary, and cognitive load under group performance constraints rather than sensory or motor access. The concrete recommendations — low-barrier keyword input, galleries of example attributes (lighting presets, camera angles) instead of textual descriptions, multi-output "pick one" reflection, visible authorship attribution — transfer directly to other AIGC tools used by older adults or low-literacy populations. The COM-B framing gives designers a way to argue for AI as a mediator of capability rather than a replacement of expertise, which is relevant to any assistive AI context. Limitations worth noting: the sample is small (15 + 16), culturally specific to urban Chinese senior-university dance communities, short-term, and does not measure real-stage deployment, long-term creative outcomes, or the cost of AIGC usage at scale. The authors acknowledge that LLM and text-to-video complexity can exceed older dancers' cognitive load, suggesting future work on complexity-adjusting language models and visually-driven (not text-driven) refinement.